Selecting random points of a PolyData (Stratified sampling)

Support
21

Thank you for the suggestion, this point distribution looks very nice - random, but uniform.

22

The algorithm is pretty simple. The process is called by some people “dart throwing”.

Shuffle the points from the input. For each point taken in the shuffled order, if no point in the output is closer than Radius to the sample, you throw it away. If not, you keep it in the output.

2 Likes
23

Will this algorithm work for meshes having around 2-3 million points ?
Currently the code is crashing and I am checking if reducing the dimension will make it work.

24

No, it didn’t work for a small mesh (or pointset).

25

I don’t know if it helps but try to check if your mesh has non-manifold edges. Maybe it doesn’t work because of that

26

I tried for pointset also i.e. not edges but it crashed.

Also checking the sphere test case https://gitlab.kitware.com/vtk/vtk/-/blob/v9.1.0/Filters/Points/Testing/Cxx/TestPoissonDiskSampler.cxx

27

If it only crashes if you have millions of points then the problem might be that you run out of memory. Please run your executable in a debugger (or get a stack trace) to see where the crash happens and what the error is.

28

Ok I will do it. But wanted to inform that the size is not a problem. It crashes for a small mesh also.
I also wanted to confirm if the input should be a mesh or pointset. As per the document, it should take pointset but I checked and the method works for a polydata sphere, however if you convert the polydata to pointset then it crashes for the same input sphere.

Following example code for demonstration:

import vtk

def readvtk(filename):
    a = vtk.vtkPolyDataReader()
    a.SetFileName(filename)
    a.Update()
    m1 = a.GetOutput()
    return m1

sphere = vtk.vtkSphereSource()
sphere.SetThetaResolution(200)
sphere.SetPhiResolution(100)
sphere.SetRadius(1.0)
sphere.Update()
sphere = sphere.GetOutput()

points = sphere.GetPoints()


pointset = vtk.vtkPointSet()
pointset.SetPoints(points)

print(pointset.GetNumberOfPoints())

f = vtk.vtkPoissonDiskSampler()

if 1:
    print('Using pointset')
    # pointset as input will fail
    f.SetInputData(pointset)
else:
    print('Using polydata')
    # polydata as input works
    f.SetInputData(sphere)

f.SetRadius(0.1)
f.Update()

output = f.GetOutput()

print(output.GetNumberOfPoints())
29

Your code snippet works well for me on Windows, using VTK v9.1 (20220125). There is no crash and prints 751.

30

It will work for polydata but not for pointset. There is a flag in the code snippet to test the two cases.

To see if it is not due to installation issues I created a Colab notebook also.
https://colab.research.google.com/drive/1EayhemN5FHRFaZMbTId7U8zozFY6FQb7?usp=sharing

Also, there is one line by mistake in the documentation which needs to be removed.
https://vtk.org/doc/nightly/html/classvtkPoissonDiskSampler.html
“generate point normals using local tangent planes”

I am also uploading a small test triangle mesh to see if vtkPoissonDiskSampler works on it.
It crashes for me.
basic_gem.vtk (866 Bytes)

31

The VTK-9.1.0 version you are using does not include the fix for the crash that happens when point normals are missing. You can either use a more recent VTK version or add normals to your points.

1 Like
32

Thank you got it to work with the latest build.
One can download the nightly build from this link (Download artifacts).

https://gitlab.kitware.com/vtk/vtk/-/jobs?page=3&scope=finished

1 Like
33

@Yohann_Bearzi We need this algorithm to give a reproducible sampling of points. Could we add a user-provided seed in the implementation?

34

That should be okay to do. Can you provide an MR, @pranjal.sahu ?